System and method for rod pump autonomous optimization without a continued use of both load cell and electric power sensor

Abstract

A method, a computer program product, and a system for pump control that incorporates software algorithms, artificial intelligence, subject matter expertise and hardware for the autonomous optimization of a rod pump in a producing oil well, including various systems. The subject of the invention that is named here The Rod Pump Surveillancer System, is a built in a Pump Controller and integrates themodels for generation and diagnostic classification of dynamometer cards, the Neural Fuzzy Logic Algorithm for a programmable logic controller functioning stand alone, or connected to an edge computer, a server at the office or in the cloud, and the program software for the Human Machine Interphase. The method includes a developed model to generate downhole dynamometer cards based on data from two sensors. A programmable logic controller and a Human Machine Interphase device is used to further enhance control capabilities.

What is claimed is: 
     
         1 . A method, a computer program product, and a system for pump control that incorporates data from fit for purpose sensors, transducers, meters, artificial intelligence tools, optimization algorithms and subject matter expertise for autonomous optimization of a rod pump producing oil well, comprising:
 a) a model to generate a Dynamometer Card based on data from two sensors, being the first one an accelerometer attached to the polished rod and the second one a positioning sensor attached to the horse head and a machine learning tool that enables a data-driven determination of the shape of the downhole dynamometer card using a database of real downhole dynamometer cards and Artificial Neural Network;   b) a model to classify Dynamometer Card based on the generated Dynamometer Card in 1a) and a Machine Learning tool that enables the diagnostic of the pump operating condition using a data base of real downhole dynamometer cards, labelled according to the prevailing operating condition as determined by a Subject or Domain Matter Expert, which may include one or a combination of two, three or of multiple operational conditions occurring at the same time during the operation of the pump and sucker rods, at a given point on time, e.g. fluid pounding only or fluid pounding and leaking standing valve or fluid pounding, leaking standing valve and pump plunger tagging up-stroke or down stroke, etc;   c) a program software for the programmable logic controller - PLC on the basis of Neural Fuzzy Logic, wherein the input data incorporates the by means of neural network generated and classified dynamometer card - 1a and 1b, measured parameters from reliable surface sensors and other calculated parameters, enabling autonomous optimization of the rod pump operation, by interacting with the variable speed drive-VSD, valve actuators, the start and stop switch, among others;   d) a Human Machine Interphase - HMI device that displays the menu comprising 5 sub-menus: Data Input, Monitor, Troubleshooting, Optimizer and Operation. It enables the users to enter the input data. Further it shows the actual and trend of the key variables that enable to monitor the operation and shows the performed diagnostic of any anomaly that may be occurring or may be about to occur. Further in the menu are the Troubleshooting module and the Optimization module, while the Operation menu is the default screen that provides an overview of the current status of the rod pump system; and   e) a computer program that is called here The Rod Pump Surveillancer - RPS System and is built in a Pump Controller that integrates a) and b) the models for generation and classification of the dynamometer cards, c) the algorithm and software program for the programmable logic controller - PLC and d) the program software for the Human Machine Interphase - HMI.   
     
     
         2 . The method of  claim 1 , wherein the Dynamometer Card Generation and Classification models use data from two sensors; an Accelerometer - attached to the polished rod and a Positioning Sensor - installed on the horse head or above the saddle bearing, which are robust devices also known as Inertial Measurement Unit - IMU sensors, where the Positioning Sensor is comprised of both an Accelerometer and a Gyroscope. The said sensors can transmit the data to the Pump Controller via electrical cable, fiberglass, electrical cable, radio or wireless. 
     
     
         3 . The method of  claim 2 , wherein both, the model that generates the dynamometer card - constructing the shape of the dynamometer card, and the model that performs the classification - predicting the type of operational condition that is occurring, utilize a neural network technique. Two versions have been implemented. One uses a machine learning model of supervised learning for applications that are executed in microcontrollers or low capacity microprocessors e.g. for local installation where there is no electrical power. The preferred embodiment uses a model developed with supervised and unsupervised deep learning as well as more robust variants of supervised and unsupervised machine learning, to be run in clusters, servers or high performance computers or CPUs at the well site. 
     
     
         4 . The method of  claim 2 , wherein to perform the generation or classification models first an updated dynamometer card is recorded using a load cell or sensor and a positioning sensor - e. g. using the Echometer tool, when the present method is run in the subject well for the first time. This card is used for calibration purposes, thereafter the models carry on generating and classifying the dynamometer cards on a continuous basis. The generation rate of dynamometric charts depends on the strokes per minute -SPM of the unit, requiring at least two complete strokes to make a good data collection. During the initial calibration process two processing options are evaluated. The first one is in a batch form that first collects a sample of data and then process them to reconstruct the dynamometric chart and classify it, while the second one is done through a time series that implies acquiring the data continuously and making predictions based on a time space of at least a couple of strokes. It is to note that the load on the surface polished rod is determined comparing the dynamometer card recorded in the calibration phase with the generated dynamometer card, as the model generates the shape of the dynamometer card, it does not calculate the load. 
     
     
         5 . The method of  claim 2 , wherein the required processing modules of the generation and classification models are described as follows: (a) The Real Time Clock - RTC, that allows to make a temporary trace to the register, for both, the classification and for the generation models. (b) The liquid-crystal display - LCD interface, that allows to view in the field the system data, such as time, date, the dynamometer card, the classification, the recommendation and the historical events of the day in the absence of a Human Machine Interphase - HMI. (c) The data acquisition module - DAQ that allows the synchronization of the request for information and the reception of data from the sensor. (d) The Data Manager, is a software module that allows managing the information (position and load), communicates with the cloud or the local processor in case the models run locally. (e) The Communication Module, e.g. the General Packet Radio Service - GRPS is a transmission module that uses the 3G cellular network to transmit and receive information from the cloud. It can transmit the raw information to be processed in the cloud or for local processing, in data packages. (f) The two Artificial Intelligence - IA Generation and Classification Models, that have been implemented in the present application, which can be executed in the cloud or locally, and are in charge of processing the information from the Data Manager, having as input the vector of acceleration and position characteristics. 
     
     
         6 . The method of  claim 2 , wherein for the supervised Machine Learning the training data set for the model to generate the dynamometer card contains as input data the time in seconds, the load on the plunger in pounds, the acceleration in units of gravity - g, the positioning in the polish rod and the position on the plunger in inches. For the classification or the diagnostic part, the input contains and the dynamometer card labelling that indicates the type of prevailing operational condition of the pump and the sucker rods, as determined by a Subject Matter Expert. The considered operational conditions that are classified as part of the diagnostic module include among others the following conditions: fluid pound, gas interference, standing valve leakage, travelling valve leakage, broken rod, stretched rod, full load production, unanchored tubing, hole in barrel, plunger tag on up-stroke, plunger tag on down-stroke, worn pump, reduced tubing diameter, among others, as well as a combination of those conditions that could occur at the same time, e.g. two or three conditions.. Further the Training Set contains the same training parameters, yet from different wells. The initialization contains initial randomly generated weighting of the network. Further for the training of the Neural Network normalized and pretreated data is utilized, and as the Loss Function the Mean Square Error is used, that indicates the accuracy with respect to the real dynamometer card. Further a Gradient Descendent Optimizer is utilized to correct the initial weighting factor in such a way that as the Epochs pass the error decreases. Finally, the termination criteria is determined either by a number of Epochs or by the stabilization at a given low error value. 
     
     
         7 . The method of  claim 2 , wherein for the supervised machine learning the training set for the model to classify the dynamometer card - a multi label classification, containsthe same features that the training of the model to generate the dynamometer card, except that the Loss Function is based on the Categorical Cross Entropy and that the termination criteria is based on achieving an accuracy above of 92%. Further the trainingprocess can also be carried out using other techniques such as supervised and unsupervised deep learning, and other techniques of recent and future development. 
     
     
         8 . The method of  claim 2 , wherein the preferred embodiment for the wells where there is no electrical power available and the use of a battery and / or solar panel is needed, incorporates a microcontroller built in a transmission device that performs the reading of both sensors from  claim 2  and performs the data pre-processing for the acceleration and the position as well as the extraction of the main characteristics, by dividing the recorded acceleration data into four blocks. From the entire register only onestroke is extracted according to the position register, this stroke is divided into four blockswith the same number of records each, from each group the main characteristics are extracted, which are inputs for both models - the Generation and the Classification Models. The information transfer to the field computer, e.g. CPU is transferred through the RS485 protocol, Modbus, or ethernet, among others. 
     
     
         9 . The method of  claim 2 , wherein the preferred embodiment for the wells where electrical power is available, the field computer, e.g. CPU can perform the reading of both sensors from  claim 2  and performs the data pre-processing for the acceleration and the position as well as the extraction of the main characteristics, by dividing the recorded acceleration data into four blocks and further feed it to the models for generation and classification of the dynamometer cards. Alternatively, more advanced programmable logic controllers - PLCs come with a CPU or microprocessor built in that can be suited toperform the above function. 
     
     
         10 . The method of  claim 2 , wherein the dynamometer card results from a data-driven generated model, another embodiment incorporates data resulted of measurement of load, carried out using a cell attached to the polished rod that includes at least an ultrasound wave device to determine the deformation of the rod and therefore the load. Alternatively, this load can also be determined via the use of a cell attached to the polished rod that incorporates at least a camera and an image processing device to determine the deformation of the rod and therefore the load. 
     
     
         11 . The method of  claim 2 , wherein both the Generation Model and the Classification Model configures an application called here - Dyna Chart App that is a system composed of hardware and software modules that allow both the generation of dynamometer cards and its automatic classification as described above. Further it also can be run on a standalone mode, without a pump controller. 
     
     
         12 . The method of  claim 1 , wherein the programmable logic controller - PLC utilizes among others a Neural Fuzzy Logic Algorithm - NFLA. It improves the diagnostic and control capabilities, based on the integration of multiple parameters that enable the proper identification of the rod pump operation anomaly cases, and the problems affecting the other subcomponents of the Integrated Production System - IPS, the inflow and the outflow, as well as the Identification of Production Improvement opportunities. 
     
     
         13 . The method of  claim 12 , wherein the input data for the Neural Fuzzy Logic Algorithm - NFLA includes the output of the generated and classified dynamometer cardusing neural network, the data recorded by reliable surface sensors and other calculatedparameters, in order to come up with specific recommendations that translate in optimized control measures, in contrast to other PLC only based solutions that have limitations withdata driven models using artificial intelligence - Al tools and rely on downhole sensors that are prone to fail or loss communication and are mainly focused on the downhole pump operation while neglecting the other subcomponents of the Integrated Production System. 
     
     
         14 . The method of  claim 1 , wherein a Human Machine Interphase - HMI device displays the menu comprising modules related to the input data, monitoring, troubleshooting, optimization and the operational default display screen. It enables the users to enter the input data of the three subsystems of the Integrated Production System - IPS for the subject well. Further it shows the actual and trend values of the key variables that enable to monitor the operation and shows the performed diagnostic of any operational condition or conditions that may be occurring or may be about to occur. Further in the menu is the Troubleshooting module that shows the recommended corrective action and the optimization module that shows the recommended action to increase oil production both are performed on autonomous mode in the preferred embodiment. 
     
     
         15 . The method of  claim 1 , wherein the computer program is called here The Rod Pump Surveillancer - RPS System and is built in a Pump Controller that integrates the models for generation and classification (or diagnostic) of the dynamometer cards, the algorithm program for the programmable logic controller - PLC, the microcontroller device, the edge computer and the program for the Human Machine Interphase - HMI. 
     
     
         16 . The method of  claim 15 , wherein, specific algorithms are used to link all the components of the RPS System: CPU or Microcontroller, PLC, HMI, sensors, meters, valve actuators, VSD, the outcome of the generated and classified dynamometer cards and the determined parameters characterizing the three subcomponents of the integrated production system IPS - the reservoir, the pump and the outflow subsystems, such as the downhole flowing pressure Pwf, the liquid flow rate QI, the oil deferment, the flow conduct diameter above the rod pump, the effective pump volume, the pump wear, among others, as opposed to other systems that are constrained to the rod pump only. 
     
     
         17 . The method of  claim 15 , wherein the hardware and software enable for ample range of application that goes from remote surveillance only to an onsite full autonomous optimization and anything in between, as required by the particular field application, and as justified by the production rate of the well. E.g. there is a configuration for low to very low rate wells and another one for high to very high rate producers. Further, the control capabilities of this application enables full autonomous pump operation by incorporating a Variable Speed Drive - VSD, flow line regulator valves and choke valves in the flow line and, or in the casing valve, wherein the choke valve can be operated by an electrical, pneumatic, or hydraulic driven actuator or adjusted manually on-site by the user, according to the recommendation of the pump controller software. 
     
     
         18 . The method of  claim 15 , wherein for low rate wells and in the absence of a Variable Speed Drive - VSD, microcontrollers or processors and a Programmable Logic Controller - PLC can be incorporated on the wellsite to stop and start the well as determined by the built-in software. Whereby the PLC can also be a conventional one, or of the type that has at least an embedded microprocessor, or CPU built in. 
     
     
         19 . The method of  claim 15 , wherein it can be used in versions for hardware based on a computer processing unit-CPU, microcontrollers, and on a programmable logic controller - PLC with a CPU (edge computer) or a microprocessor built in or embedded or a combination of them, E.g. for high rate wells. Alternatively, the software program called here The Rod Pump Surveillancer - RPS System can also be installed in a Variable Speed Drive-VSD and perform as an operating mode. 
     
     
         20 . The method of  claim 15 , wherein it can be applied for a single well or for a group of wells by incorporating a distributed control system - DCS. Moreover, all the modules of the Rod Pump Surveillance - RPS System can be used or part of it, on the wellsite, the office server or in the cloud. It also can interact with other already existing systems in the user’s facilities that perform simplified tasks such as basic alarms, start-stop functions or parameter trend display.